ISSN 2042-2695

CEP Discussion Paper No 1515

November 2017

Supplier Search and Re-Matching in Global Sourcing: Theory and Evidence from China

Fabrice Defever Christian Fischer Jens Suedekum

Abstract In this paper, we consider a dynamic search-and-matching problem of a firm with its intermediate input supplier. In our model, a headquarter currently matched with a supplier, has an interest to find and collaborate with a more efficient partner. However, supplier switching through search and re-matching is costly. Given this trade-off between the fixed costs and the expected gains from continued search, the process will stop whenever the headquarter has found a sufficiently efficient supplier. Using firm-product-level data of fresh Chinese exporters to the United States, we obtain empirical evidence in line with the predictions of our theory. In particular, we find that the share of short-term collaborations is higher in industries with more supplier-cost dispersion, an indication of higher expected search opportunities. Keywords: input sourcing, relational contracts, supplier search JEL: F23; D23; L23 This paper was produced as part of the Centre’s Trade Programme. The Centre for Economic Performance is financed by the Economic and Social Research Council. The authors thank Pol Antras, Wilhelm Kohler, Erdal Yalcin, and the participants of the CESifo Summer Institute 2015 ("New developments in Global Sourcing") in Venice for helpful comments and suggestions. All errors are our responsibility.

Fabrice Defever, City University of London, CESifo and Centre for Economic Performance, London School of Economics. Christian Fischer, University of Bayreuth, Germany. Jens Suedekum, Düsseldorf Institute for Competition Economics and CEPR. Published by Centre for Economic Performance London School of Economics and Political Science Houghton Street London WC2A 2AE All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means without the prior permission in writing of the publisher nor be issued to the public or circulated in any form other than that in which it is published. Requests for permission to reproduce any article or part of the Working Paper should be sent to the editor at the above address. F. Defever, C. Fischer and J. Suedekum, submitted 2017.

1 Introduction

Numerous studies from the management literature argue that headquarter corporations have a

vital interest in establishing and maintaining long-term collaborations with their intermediate

input suppliers.1 But before such an enduring relationship can be formed, the headquarters

first needs to find a supplier who is technologically capable of producing the desired component

in appropriate quantity and quality and at low costs. Finding such an efficient and reliable

partner is not easy, and can be thought of as a complicated search-and-matching problem that

may involve a time-consuming trial-and-error process before the firm is finally satisfied.

Several authors have recently incorporated elements of search and matching into models

of international trade. Rauch (1999) and Antras and Costinot (2011) study the search of a

domestic producer for a foreign sales agent or distributor. Once the firm has found a suitable

agent and a long-term collaboration is formed, the partners also start investing more into the

relationship. This has been shown by Araujo, Mion and Ornelas (2016) and Aeberhardt, Buono

and Fadinger (2014). In their frameworks, exports are increased at the intensive margin as the

firm builds up trust that the current foreign distributor does not behave opportunistically, cheat

on the exporting firm, engage in haggling and re-negotiation, and so on.

When it comes to global sourcing (i.e., the search-and-matching decision of the domestic

importing firm with respect to their foreign input suppliers), there is consistent empirical evi-

dence showing that firms indeed engage in a trial-and-error search. Rauch and Watson (2003)

and Besedes (2008) find that domestic firms initially place small test orders when they first deal

with a new foreign supplier. If they are dissatisfied, they terminate the collaboration, while

they increase order volumes when they are satisfied. There are also recent theoretical models

that explicitly consider the search of domestic or multinational enterprises for foreign input

suppliers, see Rauch and Watson (2003), Carballo, Ottaviano and Volpe-Martincus (2013) and

Bernard, Moxnes and Ulltveit-Moe (2014). However, until very recently, global sourcing models

had a hard time coming to grips with those empirical patterns about trial-and-error search.

A starting point of the global sourcing literature is the canonical incomplete contracts

model by Antras (2003) and Antras and Helpman (2004), which places the seminal property-

rights theory of Grossman and Hart (1986) and Hart and Moore (1990) in an international

trade context. It features hold-up and underinvestment problems as the key elements of the

buyer-supplier relationship, which can be fine-tuned by the firm’s decision how to organize the

production process and the ownership structure along the value chain. As it is well known, these

baseline models by Antras (2003) and Antras and Helpman (2004) are static models of a one-

shot interaction and, therefore, not suitable to study long-term collaborations. However, several

authors have recently embedded this model into a dynamic setting with repeated interactions;

see Baker, Gibbons and Murphy (2002). In particular, Kukharskyy (2015), Kamal and Tang

(2015) and Kukharskyy and Pfluger (2011) show that the inefficiencies caused by contract

incompleteness can be overcome by relational contracts, that is, by informal, non-binding and

purely trust-based agreements to cooperate in an enduring relationship. These papers study

1Early contributions come from Dwyer, Schurr and Oh (1987) or Kalwani and Narayandas (1995).

2

the crucial ownership choice of integration versus outsourcing in such a dynamic global sourcing

model.2 However, they consider a setting with a fixed partner, and therefore cannot capture

the evidence by Rauch and Watson (2003) and Besedes (2008) on trial-and-error search for

efficient and reliable component manufacturers.

In our own related research, see Defever, Fischer and Suedekum (2016), we have introduced

supplier search and re-matching into a dynamic property-rights model of global sourcing with

relational contracts. In our repeated-game setup, the domestic-headquarter firm observes the

efficiency of the current foreign supplier, and decides in every period whether to stick with that

partner or to engage in a costly search for another, potentially better match. Moreover, in each

round the firm can promise the current supplier a relational contract (RC) via an ex post bonus

if the supplier does not behave opportunistically, but obeys to their ex ante agreement. Much

as in Kukharskyy (2015) or in Kamal and Tang (2015), we find that such an efficient RC can

be implemented in equilibrium if agents are patient enough, and this RC then induces first-

best effort levels and overcomes the underinvestment problem. Going beyond this insight, our

model then makes clear that long-term collaborations (LTCs) and relational contracts (RCs)

are two different concepts that sharply need to be distinguished. If agents are impatient, the

firm may collaborate non-cooperatively with a supplier on a long-term basis, but they are never

able to establish an efficient informal agreement. Moreover, our model shows that some firm

only start a RC once they are satisfied with the efficiency of its current supplier. In other

words, the endogenous decision to stop searching for a better partner and, thus, to launch a

LTC causes the emergence of a RC. As this comes with increasing investment levels of both

the headquarters and the supplier, our model is therefore fully consistent with the evidence by

Rauch and Watson (2003) and Besedes (2008). More generally, our model’s message is that

the decision of with whom to interact is crucially important for the interrelated question of how

the contractual nature of this interaction will look.

The aim of this chapter is to complement the analysis by Defever et al. (2016) in some

important dimensions. In particular, we start from a stripped-down version of our model and

completely assume away contract incompleteness. The investment levels of the two partners

are fully verifiable and contractible, and hence, relational contracts are not needed to induce

efficiency in this environment. This simplification lets us focus exclusively on the firm’s search

and matching problem. The trade-off that we study in this chapter is simple: The headquarters

is currently matched with a supplier of a certain type, more specifically, with a certain unit

cost level that determines the supplier’s efficiency. The firm has an interest to collaborate

with a supplier that is as efficient as possible, because this also raises the firm’s own profits.

But supplier switching is subject to a fixed cost for search and re-matching. The firm then

essentially weighs this fixed costs against the expected gains from continued search, and will

stop searching if it has – at some point – found a supplier that is efficient enough, i.e., with

unit costs below a certain threshold level.

2There is also a more broadly related literature which has studied theoretically and empirically how rela-tional agreements can overcome hold-up problems in the context of firm organization and input sourcing, seeMacchiavello and Morjaria (2014), Board (2011) and Corts and Singh (2004).

3

While this basic formalization of the search-and-matching process is similar as in Defever

et al. (2016), we consider two important generalizations in this chapter. First, rather than

just assuming two supplier types (with high and low unit costs) as in that paper, we work here

with a general function g(c) that describes how unit costs are distributed across the mass of

potential suppliers in the economy. Second, and even more importantly, we elicit the role of

uncertainty when the firm searches for a new partner, and encounters a candidate supplier in its

random search. The firm may or may not be able to infer the efficiency of this candidate before

definitively giving up its previous match; if it is able to do so, it can separate the processes of

search and re-matching, and decide to stick with its old partner if the encountered candidate

is not more efficient than its current match. When the candidate’s efficiency is only revealed

after the first round of interaction, however, the firm will always re-match when it decides to

search, and this corresponds to Rauch’s notion of trial-and-error search for component suppliers.

It turns out that both model versions, with search and re-matching being separable or non-

separable, give rise to one specific prediction that we establish in Propositions 2 and 2’ below.

Namely, we show that the firm in our model tends to search more if the distribution g(c) is more

dispersed.3 The intuition is that, with more dispersion, more highly efficient matches are out

there in the population of candidate suppliers, which in turn makes searching more attractive.

In the final section of this chapter, we then aim to address the empirical relevance of this

novel theoretical prediction. The main challenge we are facing here is data availability. Our the-

ory is about particular matches of domestic buyers and foreign suppliers, but current data only

very rarely allows observing such matches. A notable exception is the paper by Eaton, Eslava,

Jinkins, Krizan and Tybout (2014) who are able to construct pairs of Colombian exporting

firms and their importers in the United States. They show that most exporters contract only

with a single importing partner, thus suggesting that most trade is indeed relationship-specific.

Moreover, they find considerable variation in relationship durations in the data, pointing to

a co-existence of long-term and one-shot collaborations in that market. More recent studies

which also use proprietary Census data on matches of US importers and foreign exporters

include Monarch (2015) and Kamal and Tang (2015).

In this paper, we exploit Chinese customs data that encompass all export transactions to

the United States, Japan, the United Kingdom and Germany between 2000 and 2006. We build

a sample of firms that start exporting a particular product (6-digit industry level) to one of

those destinations, but unfortunately we cannot identify the buyer. Still, following these fresh

Chinese exporters over time, we observe which firms still export the same product to the same

destination after a few years, as opposed to those that have terminated that exporting activity

in the meantime. This allows us to build a proxy for whether a particular export transaction

can be thought of as a long-term collaboration or rather a one-shot deal, and to construct the

share of short-term relationships across Chinese HS-6 industries. We correlate this share with

an industry-wide measure of cost dispersion across Chinese firms, which we construct from

unit-value data.

3More specifically, a mean-preserving spread of g(c) will increase the critical search cost level below whichthe firm decides to engage in supplier switching.

4

Our model tells us that this correlation between cost dispersion and the share of short-

term collaboration should be positive, because headquarter firms tend to search more, and will

thus exhibit more supplier turnover and one-shot deals in a given time frame. And indeed,

the empirical evidence from our Chinese data is firmly in line with this theoretical prediction.

More precisely, by aggregating the Chinese export transactions at the HS-6 product level,

we consistently find a positive correlation between the share of one-shot transactions and the

variance in the log of unit-value within an industry. This result is robust for exports across

different destination markets, for independent Chinese firms as well as for Chinese foreign

affiliates, as well as for different types of export transactions. Moreover, apart from that, this

chapter also establishes some interesting stylized facts about the resilience of Chinese exporters

that may be useful independently from the empirical test of our model.

The rest of this chapter is organized as follows. Section 2 describes our model, and Section

3 introduces the data and present the empirical results. Section 4 provides a brief conclusion.

2 The Model

2.1 Environment

Consider an industry where a single headquarter firm (labeled H) sources a customized in-

termediate input to produce a final good. There are many suppliers in that industry who

could potentially manufacture this component, and those suppliers differ in their efficiency. In

particular, we assume that there is a continuum of potential suppliers, which are all identical

except for the constant marginal costs c that they would incur in the production of the input.

Costs are distributed across potential suppliers according to some probability density function

g(c) with support [c, c], where c > 0. The corresponding cumulative density function G(c) is

continuously differentiable in c.

The headquarters initially gets matched with one supplier M0 with unit costs c0 randomly

drawn from g(c); this initial matching is costless. From the production and sale of the final

output, the firm can appropriate a flow payoff π(c) per period that is positive, differentiable

and monotonically increasing in the supplier’s efficiency, i.e., decreasing in marginal production

costs: ∂π∂c< 0. Production is repeated ad infinitum, and the discrete time periods are indexed

by t = 0, 1, 2, ..., and δ > 0 denotes the time discount factor.

In this model, the firm has an incentive to collaborate with a supplier that is as efficient

as possible (with the lowest unit costs c), and this raises the static payoff per period as well

as the discounted stream of future profits. Yet we assume that a switch of suppliers involves a

fixed cost. More specifically, we start our analysis with the case where the firm can perfectly

infer the efficiency of candidate suppliers, which allows it to separate the processes of search

and re-matching. Every stage of our infinitely repeated interaction then has the following two

consecutive steps:

1. Transaction stage: The current supplier delivers the input, final output is produced

5

and sold, and the headquarter firm receives the payoff π(c), with ∂π/∂c < 0.

2. Re-matching stage: The firm can decide to search for a new supplier. When deciding

to search, it incurs a fixed cost F > 0. Let ct be the unit cost level of its current supplier,

and ct+1 the unit cost of the new supplier that it has encountered during its search.

The cost ct+1 is randomly drawn from the distribution function G(c) which is i.i.d. over

periods. The firm can perfectly observe ct+1 before deciding to re-match. If the cost draw

is such that ct+1 < ct, the headquarters re-matches and continues the game with the new

supplier. If ct+1 ≥ ct it keeps its previous supplier.

Some comments about this simple setting are in order before we continue with the analysis.

First, notice that our model is a stripped-down version of Defever, Fischer and Suedekum

(2016). We abstract entirely from contractual frictions and focus solely on the firm’s search

and matching problem in an environment where, within a given match, investment choices are

perfectly observable and contractible. To keep the exposition as parsimonious as possible, we

furthermore stick to a reduced-form model that does not explicitly specify market demand,

investment decisions, and revenue sharing. Those elements could be easily introduced, but

they are not necessary to derive the main theoretical predictions that we later take to the data.

Furthermore, our model focuses entirely on the decisions of the headquarter firm H but it

assigns a passive role to the suppliers. The firm, which cannot be in a relationship with more

than one supplier at the same time, seeks for the best possible partner but the supplier itself

(regardless of c) has no incentive to break up the collaboration but stays inside the relationship

forever unless it is fired by the headquarters. We make this assumption, which could be justified

by a zero (or sufficiently small) outside option for all potential suppliers, for convenience in order

to focus on the re-matching incentives of the headquarter firm. In addition, this assumption

allows us to abstract from the strategic incentives of the supplier. For this reason, we also

do not need to specify the payoff for the supplier, but we simply assume that it is positive

and, thus, better than the outside option. We also abstract from break-up costs, but one may

interpret the search costs F at least partly as a compensation payment for the old supplier.

Finally, in the baseline version of our model studied in Section 2.2, we assume that the firm

has the option to stick to its old supplier if it is not satisfied with the candidate supplier that

it encounters during the search in any given period. On the one hand, this is realistic since

suppliers have small (zero) outside options and would therefore always prefer to stay inside the

relationship, even if the firm actively seeks for a better partner. On the other hand, there are

also arguments in favor of an alternative assumption, namely that the firm must re-match if it

decides to engage in searching. Specifically, this alternative scenario can be understood in terms

of the information that is available to the firm about the candidate that it encounters in the

search process. In the baseline version, the firm can perfectly observe its cost-efficiency before

actually re-matching to it. This allows the firm to separate the steps of search and re-matching

into two distinct decisions. In the second scenario, however, we may assume that the firm can

only observe the costs of the newly drawn supplier ex post, i.e., only after the first period of

6

collaboration. When the candidate’s costs are uncertain, the firm will always re-match when

engaging in search (otherwise it would not pay F ), and the search and re-matching decisions

then become non-separable. We study this alternative setting below in Secion 2.3.

2.2 Separable search and re-matching

Starting with the first scenario, denote by c the cost realization of the firm’s current supplier,

and by π the firm’s associated payoff per period. In order to see whether the firm engages

in supplier search, given that c0 = c, the expected payoff when engaging in supplier search,

denoted by E[πS | c0 = c], has to be larger than the continued payoffs with the current match,

which is given by the infinite geometric series of the per-period payoff π . This can be expressed

formally by the following supplier search incentive constraint:

E[πS | c0 = c] >π

1− δ(IC-Search)

Suppose it is profitable for the headquarters to search in period 0, i.e. (IC-Search) is fulfilled.

It then pays F and with probability 1 − G(c) draws a supplier that does not give it a payoff

improvement when compared to the initial match. If that happens, it will stick to the initial

supplier and search again at the end of the following round, because search must be optimal

at the end of round 1 if it was optimal at the end of round 0 in this infinitely repeated game.

With probability G(c), however, it finds a supplier with lower costs and, hence, re-matches. In

the period after re-matching, the firm re-evaluates (IC-Search) with the cost-realization of the

new supplier and thereby determines whether it will engage in further search for an even more

efficient partner, and so on.

The expected search payoff E[πS | c0 = c] can be formally expressed as the solution to the

following program:

V0 = π − F + δV1

Vt = Pr (c < c)E[π | c < c]

1− δ+ Pr (c ≥ c) (π − F + δVt+1) , t = 1, 2, ...

The first equation captures the value of the relationship at time 0, denoted V0, which includes

the option value of searching in period 1. This search, formalized by the second equation, leads

to a payoff improvement (and, thus, to an actual re-matching) only with a certain probability.

With the respective counter-probability, however, the search is unsuccessful in period 1, in

which case it continues in period 2, and so on.

It is important to understand that the expression (IC-Search) is conditional on the current

cost realization, i.e., a new version of it has to be formulated for every supplier that the firm

encounters in the sequence of re-matches that it performs. Then, observing that the decision

problem is the same in every round where the headquarters is still matched with c and hence is

on the search for a better partner, we can set Vt = Vt+1 for t = 1, 2, ... and simplify the program

7

to:

V0 = π − F + δV1

V1 =1

1− δ(1−G(c))

[(1−G(c))(π − F ) +

G(c)E[π | c < c]

1− δ

](1)

Using (1) in (IC-Search), we can then solve for F and obtain a search cost threshold F (c)

that determines whether or not the headquarter firm will engage in supplier search:

F <δ

1− δG(c) [E[π | c < c]− π] ≡ F (c) (2)

Expression (2) demonstrates that the fixed costs F must be low enough in order for search to

occur. Notice that the critical search cost level F (c) depends positively on δ, i.e., search is more

attractive the more patient agents are. The intuition is that future payoffs matter more, and

hence it becomes more important to find an efficient supplier with low costs. Moreover, F (c)

depends positively on G(c) and on the term [E[π | c < c]− π]. This shows that search is more

likely the higher the chance to find a more efficient supplier, and the larger the headquarters’

expected per-period payoff difference when comparing a better with the current supplier.

We summarize the result in the following proposition:

Proposition 1. Suppose the headquarters is currently matched with a supplier with marginal

input production costs c. Then the headquarters will engage in the search for a better supplier

whenever F < F (c) and stick to the current match otherwise.

Note that whenever (IC-Search) is fulfilled, the interaction of the firm with its current

supplier is short-term in the sense that it will end in the first period where the firm draws a

supplier with c′ < c. To be precise, this means that, while being engaged in search, the firm

will be matched with the current supplier for only a finite number of periods. The relationship

with a particular supplier will only be long-term (i.e. durable in all future periods) when the

probability of finding a better partner together with the size of possible payoff improvements,

does not justify the per-period search expenses, which is the case precisely when (IC-Search)

does not hold anymore.

Essentially, the search decision is driven by the comparison of search costs and expected

gains from search. The latter of which hinge crucially on the properties of the distribution g(c),

and on where in the distribution the current supplier is located. In the following, we derive

an important result that relates the firm’s propensity to search with the cost distribution of

potential suppliers:

Proposition 2. A mean-preserving spread (MPS) on the distribution function G(c) of potential

suppliers increases the firm’s propensity to search for a more cost-efficient supplier by increasing

the critical search cost level F (c) for all possible values of c.

8

To prove Proposition 2 formally, let G(c) be a MPS of G(c), and let g(c) be its associated

probability density function. Denote by F the search cost threshold resulting under G(c) and

by F that for G(c). Then for any c0 = c it must hold that

F (c) > F (c). (3)

Substituting (2) into (3) and simplifying gives

G(c) (EG[π | c < c]− π) > G(c) (EG[π | c < c]− π) .

Plugging in the conditional expected values and simplifying gives∫ c

c

π(c)g(c) dc− G(c)π >

∫ c

c

π(c)g(c) dc−G(c)π.

Integrating by parts allows us to simplify further:∫ c

c

∂π

∂cG(c) dc >

∫ c

c

∂π

∂cG(c) dc. (4)

Because ∂π∂c< 0, it follows from the definition of the MPS that expression (4) is true for any

value of c. From this we can conclude that a more dispersed cost distribution in an industry

increases the headquarters’ search and re-matching propensity.

Intuitively, a MPS can be interpreted as a shift of probability mass towards the ends of

the probability density function g(c). With relatively more probability mass close to the most

efficient supplier at c, both G(c) as well as E[π | c < c] increase which in turn shifts F (c)

upwards. This makes it more attractive to search for a partner with lower costs, essentially

because the firm only re-matches when it can improve its per-period payoff. An MPS adds

efficient suppliers to the tail of the distribution an generates such a payoff increase.

2.3 Non-separable search and re-matching

In the previous subsection, we have assumed that the headquarters can perfectly observe the

cost-efficiency of the candidate supplier before actually starting a relationship with it. Obvi-

ously, this assumption is a strong one. We now study the alternative scenario where the firm

only learns a supplier’s cost-efficiency once it is matched to it (while still having perfect infor-

mation on the distribution G(c)). Whenever the headquarters cannot observe the cost-level of

the suppliers that it encounters during the search, it will re-match to the new supplier whenever

it decides to engage in search. Otherwise the firm could refrain from searching in the first place,

and save the fixed costs F . We reformulate step 2 of the stage game as follows:

2’. Re-matching stage: The firm decides whether to re-match with a new supplier. When

deciding to re-match, it incurs a fixed cost F > 0. Let ct be the unit cost level of its

9

current supplier, and ct+1 the unit cost of the new supplier that it has encountered during

its search. The cost ct+1 is randomly drawn from the distribution function G(c) which

is i.i.d. over periods. The firm cannot observe ct+1 before the start of a transaction

relationship and hence re-matches whenever it searches.

We will first assess the robustness of Proposition 1 for this case where search and re-matching

cannot be separated. Again, we denote by c and π the current supplier’s costs and the firm’s

associated payoffs, respectively. Equivalently to (IC-Search), the firm trades off the expected

payoffs when engaging in re-matching with the continuation payoff resulting from its current

relationship. Formally:

E[πRM | c0 = c] >π

1− δ(IC-ReMatch)

E[πRM | c0 = c] follows the same logic as E[πS | c0 = c] from above, but differs in one

respect. Suppose that, after paying the search costs F , the firm draws a supplier with c > c,

which happens with probability 1 − G(c). The firm cannot observe these costs directly, but

only when arriving at the transaction stage of the following period; thus, from an ex-ante

perspective, the firm will receive the expected payoff with this worse supplier, E[π | c ≥ c], for

one period and then continue searching. Formally, then, we can then express E[πRM | c0 = c]

as the solution to the following system:

V0 = π − F + δV1

Vt = Pr(c < c)E[π | c < c]

1− δ+ Pr(c ≥ c) (E[π | c ≥ c]− F + δVt) , t = 1, 2, ...

We can bring the program into the non-recursive form:

V0 = π − F + δV1

V1 =1

1− δ(1−G(c))

[(1−G(c))(E[π | c ≥ c]− F ) +

G(c)E[π | c < c]

1− δ

]By a method analogous to what is above, we can now solve (IC-ReMatch) for F and obtain

a threshold level F (c) that determines whether the firm will engage in supplier re-matching or

not:

F <δ

1− δ

[(E[π]− π

)− (δ(1−G(c))

(E[π | c > c]− π

)]≡ F (c) (5)

Expression (5) demonstrates that search costs F have to be low enough in order for re-

matching to occur. Again, F (c) depends positively on the patience level δ. Notice that because

(E[π | c > c]− π) is always negative, (E[π] > π) is a sufficient condition for F (c) > 0; that is,

having a below-average supplier is always enough for re-matching to occur if search costs are

small enough.

We summarize the result in an adjusted version of Proposition 1:

10

Proposition 1’. Suppose the headquarters is currently matched with a supplier with marginal

input production costs c. When the headquarters can infer the cost-efficiency of a candidate

supplier only ex-post to contracting, it will engage in re-matching whenever F < F (c) and stick

to the current match otherwise.

Note that, whenever (IC-ReMatch) is fulfilled, the relationship with the current supplier

is short-term as it was also the case in the previous subsection when search and re-matching

decisions could be separated by the firm. In the present setting, however, the relationship will

definitely exist for only one period of the game, while in the previous case, the current firm-

supplier pairing would last until a more efficient supplier was found. Finally, when encountering

a better supplier in the re-matching sequence with c′ < c for which F ≥ F (c′) turns out to

hold, the re-matching process stops and the firm will engage in a long-term relationship with

this supplier.

Turning to Proposition 2, we now study again the effects of an increase in dispersion of the

distribution g(c) on the firm’s propensity to re-match:

Proposition 2’. Consider the case where the candidate supplier’s cost-efficiency is only ob-

servable ex-post to the firm. A mean-preserving spread G(c) on the distribution function G(c)

increases the firm’s propensity to engage in re-matching by increasing the critical search cost

level F (c) for all possible values of c.

The proof of Proposition 2’ goes along the same lines as in the previous case. Let G(c) be

an MPS of G(c), and g(c) its associated probability density function. Denote by F the search

cost threshold resulting under G(c) and by F the search cost threshold for G(c). Then for

Proposition 2’ to be true, it must hold for any c0 = c that

F (c) > F (c). (6)

Substituting (5) into (6) and simplifying gives:

(1− G(c)(EG[π | c > c]− π

)< (1−G(c)

(EG[π | c > c]− π

)We now plug in the conditional expected values and simplify:∫ c

c

π(c)g(c) dc− (1− G(c))π <

∫ c

c

π(c)g(c) dc− (1−G(c))π

Integrating by parts we arrive at:∫ c

c

∂π

∂cG(c) dc >

∫ c

c

∂π

∂cG(c) dc (7)

Because G(c) is a MPS of G(c), and ∂π∂c< 0, it follows that expression (7) is true for all values

of c. This concludes the proof.

11

2.4 Summary of theoretical results and main prediction

Summing up, the two model versions – with search and re-matching being separable or non-

separable – yield qualitatively similar results and predictions. In order for search/re-matching

to occur, the fixed costs F must be below a threshold level. This cutoff is stricter (the threshold

for F is lower), the lower is the unit cost of the current supplier. That is, the better the current

supplier is already, the lower is the chance that the firm will continue looking for a better match.

Moreover – turning to the key prediction of our model – both model versions show that more

cost dispersion among the potential suppliers in the industry lowers this search threshold, and

thus leads to more search/re-matching, because it leads to an expected gain from searching as

better potential matches are now out there in the industry.

Figure 1 gives a graphical summary. Making the additional structural assumption that∂2π∂c2

< 0, we can draw F as a function that is increasing and convex in c. Then, the model

predicts that for all (c, F )-combinations that lie below this function the firm will engage in

search for a better supplier and hence be in a short-term collaboration (STC) with its current

partner (blue area of the Figure). Otherwise, for all (c, F )-combinations above that convex

curve, the firm will not search, but will establish a long-term collaboration (LTC) with its

current match. Performing an MPS on the cost distribution g(c) shifts the F -function upwards

to F , and thereby enlarges the parameter range for which the firm will engage in supplier search

/ re-matching.

c

F

cc0

δ1−δ (E[π]− π(c))

LTC

STC

MPS

F

F

Figure 1: Illustration of search threshold and MPS

Our results do not differ qualitatively between the two cases, which underscores that our

predictions are robust with respect to the quality with which information on supplier costs is

transmitted to the headquarters. In particular, it indicates that the prediction that a larger

cost dispersion within an industry leads to a higher search and re-matching propensity is not

12

crucially influenced by how quickly firms learn about the supplier’s efficiency. This gives us

confidence in our industry-level prediction of a positive relationship between the cost-dispersion

of potential suppliers and firms’ propensity to engage in supplier search and re-matching.

3 Empirical Analysis

In this section, we describe how we construct proxies for the two variables that are required

to empirically test the key prediction of our theoretical model: i) whether a buyer-supplier

match is an STC as opposed to an LTC, and ii) the industry-wide cost dispersion of potential

suppliers.

3.1 Data and variables

Export transactions of fresh Chinese suppliers: The empirical evaluation of the model

exploits transaction-level customs data from the Chinese General Administration of Customs

for the period 2000-2006. For each year, the data allow us to identify the manufacturing

export transactions of Chinese firms, where an export transaction is defined as a firm-product-

destination combination. We drop all transactions with zero or negative values as well as

export transactions with destination China. Moreover, to ensure consistency of the product

categorization over time, we use the conversion table from UN Comtrade and convert the

product code used for the years 2000 and 2001 into the HS 2002 codes.

In our theoretical framework, the headquarter deals with a new supplier when it decides

to re-match. To reflect this, we only consider transactions of fresh Chinese exporters that

establish their first exporting activity, so that foreign importing firms cannot infer any quality

signals from their previous exporting experiences. Moreover, we restrict our sample to new

transactions realized in the United States, Japan, United Kingdom and Germany, which are

the top four main destinations for Chinese exports besides Hong-Kong and Macao. Specifically,

our sample is composed of Chinese firms i that export a product j to a country c in year t,

where t = {2001, 2002, 2003}, but that have not exported anything to any destination outside

China during the previous years since 2000. We exclude all industries j with fewer than ten

active firms, and those where control variables are not available.

An important observation, as shown in the upper panel of table 1, is that quite a few of

the new Chinese exporters in our data are owned by multinational corporations. This includes

Chinese foreign affiliates that are dependent subsidiaries or part of a joint venture. Altogether,

the final sample includes 28,370 fresh Chinese exporters (15,017 domestic firms and 13,353

foreign affiliates), starting a total of 166,032 new export transactions, and spans 1,192 different

HS-6-digit product categories. In our baseline estimations we include all these observations,

but we also conduct robustness checks that focus only on independent Chinese exporters.

Regarding the destination markets, while our full sample pools transaction with the United

States, Japan, the United Kingdom and Germany, we also present results for each market

independently, and the lower panels of table 1 show the number of Chinese exporters, exported

13

products and export transactions in those four markets. In addition, for our baseline estimations

we keep all types of goods, and in the robustness checks we limit this sample to complex goods

with a relatively high degree of relationship-specificity, as this might better reflect the context

of our theoretical model.

Table 1: Firms, products and destinations

All destinationsDomestic firms Foreign affiliates

# firms 15017 13353# products 1192 1166# transactions 122707 43325

All firmsUS Japan

# firms 16150 13683# products 1004 861# transactions 63564 48877

All firmsUK Germany

# firms 5996 6606# products 452 478# transactions 16531 16427

Supplier turnover: Following the fresh Chinese exporters over time, we observe which of

the firms that started an exporting activity in country c and product j in year t, terminate

their export activity in year t+ 3, as opposed to those that still export the same product j to

the same country c after three years. Ideally, we would like to observe whether the Chinese

exporter still deals with the same importer, but the data do not allow us to do so. Nevertheless,

looking specifically at the United States as the country of destination in table 3, our duration

measure provides a pattern very similar to the one observed by Monarch (2015), who is able to

use confidential US Customs data on US import transactions from China, where firms on both

sides are uniquely identified. He finds that 45% of US importers change their Chinese partner

from one year to the next. This number is quite close to the one obtained in our data (43.4%).

Aggregating to the HS-6 product level, we compute the share of transactions in industry j

that got started in t = {2001, 2002, 2003} and that have stopped in year (t + 3). This gives

us our measure on the supplier turnover in industry j, which we use as our main explanatory

variable: the share of short-term transactions.

Industry-wide cost dispersion: To construct an empirical proxy for the industry-wide

cost dispersion, we approximate the transaction-specific marginal cost of the respective supplier

by the unit-value of each transaction. The cost dispersion measure for industry j is then given

by the standard deviation of the log of unit-values within the respective HS-6 industry during

our period of observation. As usual, marginal costs are difficult to observe directly. Thus, our

14

underlying assumption, follows standard practise and postulates that Chinese suppliers engage

in fixed-markup pricing, so that the variation in unit-values captures cost dispersion. The

recent findings by Monarch (2015) support this approach. Using confidential customs data,

he shows that US importers paying the highest prices to their Chinese suppliers (also proxied

by the unit-value in his paper) are most likely to change their partner in the future. This is

consistent with Proposition 1 of our model, where re-matching is more likely the higher the

marginal costs of the current supplier.

Manova and Zhang (2012) argue that differences in unit-values could also reflects quality

variations. In this case, our empirical proxy would be capturing heterogeneity in suppliers’

quality. However, under iso-elastic demand and monopolistic competition, marginal costs and

quality differences are isomorphic in the sense that they enter equilibrium firm revenue in

exactly the same way (see Melitz and Redding, 2014). As a result, while a lower quality would

have the opposite effect on re-matching compared to lower cost, decreasing the variance of either

quality or cost would have the same impact on the decrease of supplier turnover. Hence, the

interpretation of our results would be robust to the way we interpret differences in unit-values.

3.2 Descriptive evidence

Table 2 reports the duration of the transactions for independent firms and for foreign affiliates.

It is important to take these ownership structures into account, as Chinese foreign affiliates

may differ systematically in their economic behavior from independent contractors. While the

ownership dimension does not feature in our theoretical model, the key trade-off formalized by

our model is – in principle – not limited to arm’s-length outsourcing relationships, but can also

arise with a similar intuition in the context of intra-firm trade.

Looking at table 2, of the 122,707 transactions made by domestically-owned firms, it turns

out that 59,361 (48.4%) are terminated after less than one year, while 34,175 (27.9%) of them

endure for more than three years. While transactions involving foreign affiliates tend to be

more resilient, still 28.2% of the transactions stop after the first year. Digging deeper into the

different countries of destinations (the United States, Japan, the United Kingdom and Germany

), table 3 shows that the share of one-year transactions within these four subgroups is rather

similar across destinations (from 40.4% for Japan to 45.8% for the United Kingdom).

Table 2: Duration of export transactions by ownership

Duration of Domestic firms Foreign affiliatestransactions # transactions Percentage # transactions Percentage< 1 year 59,361 48.4% 12,201 28.2%> 1 year 17,970 14.6% 5,814 13.4%> 2 years 11,201 9.1% 4,512 10.4%> 3 years 34,175 27.9% 20,798 48.0%Total 122,707 100.0% 43,325 100.0%

15

Table 3: Duration of export transactions by export destination

Duration of Exports to the US Exports to Japantransactions # transactions Percentage # transactions Percentage< 1 year 27,572 43.4% 19,764 40.4%> 1 year 9,498 14.9% 6,963 14.2%> 2 years 6,108 9.6% 4,759 9.7%> 3 years 20,402 32.1% 17,399 35.6%Total 63,580 100.0% 48,885 100.0%

Duration of Exports to the UK Exports to Germanytransactions # transactions Percentage # transactions Percentage< 1 year 7,577 45.8% 7,068 43.0%> 1 year 2,361 14.3% 2,231 13.6%> 2 years 1,477 8.9% 1,534 9.3%> 3 years 5,118 31.0% 5,594 34.1%Total 16,533 100.0% 16,427 100.0%

Ordinary least squares (OLS) regressions. In table 4, we present some OLS results at

the industry level. To do this, we first aggregate the data to the HS-6 product-destination

level. Specifically, we use all new transactions by fresh Chinese exporting firms i in industry j

to country c. We create a dummy variable indicating whether the transaction is a short-term

transaction (value 1) or if it is still observed in t+3 (value 0), and may thus be classified as LTC.

We then obtain our dependent variable by aggregating the data at the product-destination level

by computing the share of short-term transactions among all the transactions within an HS-6

product. All our baseline regressions include a set of destination fixed effects δc.

Share short-term transactionsjc = a0 + a1Cost Dispersionjc + controlsj + δc + εjc

In column 1, we regress our dependent variable on the within-product standard deviation

of the log of unit prices, while controlling for the ownership status of the Chinese firm. We

find a positive correlation between the share of short-term transactions and the within-industry

spread in unit value, as expected from Proposition 2 of our theoretical model. In column 2, we

introduce a set of additional industry-level control variables, namely the capital-labor ratio, the

human capital intensity, and the R&D-sales ratio, which are standard measures for headquarter-

intensity. Moreover, we include the relationship-specificity index by Nunn (2007), and add

demand elasticities from Broda and Weinstein (2006).4 Adding these control variable leaves

the main coefficient unchanged.

In Columns 3 and 4 of table 4, we exclude all the export transactions of foreign affiliates

before aggregating the data at the product level. This leads to a decreasing number of industries

4As data are disaggregated at the HS 10 digit level, we use the median value for each HS-6 product.

16

as we still exclude all products with fewer than ten observations. The main coefficient remains

very similar and remains significant at the 1% level. Overall, this indicates that our results

remain robust also in a smaller sample consisting only of independent Chinese-owned component

manufacturers.

Table 4: Relational-contracts and long-term collaborations: Transaction-level

Dependent variable: Full sample ExcludingShare of one-shot transactions foreign affiliatesLn unit value SD 0.050a 0.072a 0.065a 0.065a

(0.009) (0.010) (0.010) (0.011)Foreign ownership -0.207a -0.199a

(0.024) (0.023)ln capital/worker 0.001 -0.021b

(0.009) (0.010)ln skilled/worker 0.062a -0.010

(0.021) (0.026)ln R&D sales -0.009b -0.003

(0.004) (0.005)Nunn measure 0.034 0.065

(0.042) (0.056)ln demand elasticity 0.007 -0.000

(0.007) (0.008)Country fixed effects Yes Yes Yes YesObservations 3,456 3,456 1,029 1,0290R2 0.058 0.074 0.042 0.053

Notes: Dependent variable: Share of export transactions which are one-shot on the overall number oftransactions within an HS-6 product. Robust standard-errors in brackets using clustered standard errorsat the HS-6 level.

Next, in table 5 we report results separately for transactions realized in each of the top-four

export destinations, that is the United States, Japan, the United Kingdom and Germany. Our

main coefficient of interest remains stable in all specifications, and is always significant at the

1% level, regardless of whether we include or exclude foreign affiliates or whether we consider

the additional industry-level control variables.

We find that, taken together, the results in tables 4 and 5 are firmly in line with the predic-

tions of our theoretical model– more specifically, with Proposition 2, which states that higher

cost dispersion among the suppliers leads to more search and re-matching in that industry, and

thus negatively affects observed match durations.

3.3 Robustness checks and discussion

Table 6 provides a battery of robustness checks. The first one consists of dropping all trans-

actions that include processing trade. As shown in Defever, Fischer and Suedekum (2016),

processing transactions are usually associated with high buyer-supplier interrelationship and

17

Table 5: Relational-contracts and long-term collaborations: Industry-level

Dependent variable: Country of destinationShare of one-shot transactions US Japanln unit value SD 0.050a 0.065a 0.049a 0.088a

(0.009) (0.010) (0.010) (0.011)Foreign ownership -0.238a -0.235a -0.151a -0.194a

(0.034) (0.033) (0.029) (0.028)ln capital/worker -0.004 -0.043a

(0.010) (0.011)ln skilled/worker 0.030 0.117a

(0.022) (0.029)ln R&D sales -0.008c -0.010c

(0.004) (0.005)Nunn measure -0.047 0.119b

(0.058) (0.058)ln demand elasticity 0.007 0.010

(0.008) (0.008)Observations 1,004 1,004 861 861R2 0.063 0.077 0.051 0.129

Dependent variable: Country of destinationShare of one-shot transactions UK Germanyln unit value SD 0.047b 0.058a 0.036b 0.062a

(0.018) (0.021) (0.016) (0.018)Foreign ownership -0.258a -0.232a -0.187a -0.159a

(0.045) (0.047) (0.049) (0.048)ln capital/worker 0.035b 0.012

(0.017) (0.014)ln skilled/worker 0.018 0.118a

(0.033) (0.033)ln R&D sales -0.008 0.002

(0.007) (0.007)Nunn measure -0.151 0.005

(0.098) (0.106)ln demand elasticity 0.013 0.004

(0.017) (0.012)Observations 452 452 478 478R2 0.075 0.098 0.036 0.064

Notes: Dependent variable: Share of export transactions which are one-shot on the overall number oftransactions within an HS-6 product. Robust standard-errors in brackets using clustered standard errorsat the HS-6 level.

relational contracts. Columns 1 shows that the main coefficient for the within-sector price

dispersion remains extremely similar to our baseline results.

In column 2 we only keep industries where the Nunn-measure is above the median, and in

column 3 we limit the sample even further to industries in the upper quartile of specificity.

In fact, our baseline estimates are for all 1,004 product categories, some of which are rather

18

Table 6: Relational-contracts and long-term collaborations: Transaction-level

Dependent variable: Drop Nunn measure Nunn measure MainShare of one-shot transactions Processing > median > 75 percentile sectorOne shot dummy 0.069a 0.067a 0.045a 0.068a

(0.010) (0.012) (0.015) (0.013)Foreign ownership -0.123a -0.253a -0.263a -0.062b

(0.037) (0.041) (0.056) (0.029)ln capital/worker -0.023b 0.016 0.010 -0.011

(0.009) (0.015) (0.021) (0.012)ln skilled/worker 0.024 0.048 0.018 0.022

(0.025) (0.031) (0.040) (0.031)ln R&D sales -0.014a -0.012b -0.021b -0.022a

(0.005) (0.006) (0.010) (0.006)Nunn measure 0.058 2.586a 4.727a 0.064

(0.050) (0.678) (1.071) (0.059)ln demand elasticity 0.011 0.016c 0.017 0.025a

(0.007) (0.009) (0.011) (0.009)Observations 1,101 602 380 904R2 0.064 0.143 0.142 0.068

Notes: Dependent variable: Share of export transactions which are one-shot on the overall number oftransactions within an HS-6 product. Robust standard-errors in brackets using clustered standard errorsat the HS-6 level.

standardized goods or final products. Since our model features the duration of a relationship

between buyer and supplier when producing a specific intermediate input, we now limit the

sample and – rather than just adding the Nunn measure as a control variable – focus on

product categories with a high degree of relationship-specificity. As can be seen, all results

remain qualitatively unchanged, even if the point estimates become slightly smaller in Column

3. This, even when focussing on highly relationship-specific goods, we find that the within-

sectoral cost-spread increases the proportion of short-term transactions.

In column 4, since many firms export multiple products, we exclude all the transactions

that are not part of the firm’s main sector of activity which we define by the 2-digit HS product

for which we observe most exported HS-6 products. We only keep the transactions that belong

to this main 2-digit sector of each firm, and again we find that the coefficients are very similar

to our baseline estimates.

We have also conducted all these robustness checks separately for the different destination

markets, along the lines of table 5, and we have found that the pattern remains very stable in all

these specifications. That is, in line with Proposition 2 of our model, we find a robust positive

correlation between the industry-wide measure of cost dispersion and the share of short-term

export collaborations in that industry, indicating that headquarters tend to engage in more

supplier search when there is more cost dispersion among potential suppliers.

19

4 Conclusion

In this paper, we develop a dynamic search and matching model of global sourcing where a

domestic headquarters seeks to obtain an intermediate input from a foreign input supplier.

We show that the option to search and re-match with a new supplier crucially depends on

an industry’s characteristics. More precisely, the industry-wide cost dispersion of potential

suppliers affects the expected search payoff. In industries with a large cost dispersion among

suppliers, headquarters tend to search more, and will thus exhibit more supplier turnover and

one-shot deals in a given time frame. This is true both in the case where search and re-matching

are separable, and when they are non-separable.

This key prediction of our model is empirically supported in the context of export transac-

tions of fresh Chinese suppliers to the United States, Japan, the United Kingdom and Germany,

the top four main destinations for Chinese exports besides Hong-Kong and Macao. By aggre-

gating the Chinese export transactions at the HS-6 product level, we show that the share of

short-term transactions increase with an within-industry cost dispersion, which we construct

from unit-value data. This result is robust for exports across different destination markets, for

independent Chinese firms, as well as for Chinese foreign affiliates.

More generally, this chapter opens several promising avenues for future research that have

not been explored yet, in particular by documenting novel stylized facts about the resilience of

product-destination export transactions.

References

[1] Aeberhardt, R., Buono, I., and Fadinger, H. Learning, Incomplete Contracts and

Export Dynamics: Theory and Evidence from French Firms. European Economic Review

68 (2014), 219–249.

[2] Antras, P. Firms, Contracts, and Trade Structure. The Quarterly Journal of Economics

118, 4 (2003), 31–77.

[3] Antras, P. Global Production: Firms, Contracts and Trade Structure. Princeton Uni-

versity Press, 2015.

[4] Antras, P., and Costinot, A. Intermediated Trade. The Quarterly Journal of Eco-

nomics 126 (2011), 1319–1374.

[5] Antras, P., and Helpman, E. Global Sourcing. Journal of Political Economy 112, 3

(2004), 552–580.

[6] Araujo, L., Mion, G., and Ornelas, E. Institutions and Export Dynamics. Journal

of International Economics 98 (2016), 2–20.

[7] Baker, G., Gibbons, R., and Murphy, K. J. Relational Contracts and the Theory

of the Firm. The Quarterly Journal of Economics 117, 1 (2002), 39–84.

20

[8] Bernard, A. B., Moxnes, A., and Ulltveit-Moe, K. H. Two-Sided Heterogeneity

and Trade. Working Paper 20136, National Bureau of Economic Research, May 2014.

[9] Besedes, T. A Search Cost Perspective on Formation and Duration of Trade. Review of

International Economics 16, 5 (2008), 835–849.

[10] Board, S. Relational Contracts and the Value of Loyalty. American Economic Review

101, 7 (2011), 3349–67.

[11] Broda, C., and Weinstein, D. E. Globalization and the Gains from Variety. The

Quarterly Journal of Economics 121, 2 (2006), 541–585.

[12] Carballo, J., Ottaviano, G., and Volpe-Martincus, C. The Buyer Margins of

Firms’ Exports. Discussion Paper 1234, CEP, 2013.

[13] Corts, K. S., and Singh, J. The Effect of Repeated Interaction on Contract Choice:

Evidence from Offshore Drilling. Journal of Law, Economics, and Organization 20, 1

(2004), 230–260.

[14] Defever, F., Fischer, C., and Suedekum, J. Relational contracts and supplier

turnover in the global economy. Journal of International Economics 103 (2016), 147–165.

[15] Dwyer, R. F., Schurr, P. H., and Oh, S. Developing Buyer-Seller Relationships.

The Journal of Marketing 51, 2 (1987), 11–27.

[16] Eaton, J., Eslava, M., Jinkins, D., Krizan, C. J., and Tybout, J. A Search and

Learning Model of Export Dynamics. Working Paper, 2014.

[17] Grossman, S. J., and Hart, O. D. The Costs and Benefits of Ownership: A Theory of

Vertical and Lateral Integration. The Journal of Political Economy 94, 4 (1986), 691–719.

[18] Hart, O. D., and Moore, J. Property Rights and the Nature of the Firm. Journal of

Political Economy 98, 6 (1990), 1119–58.

[19] Kalwani, M. U., and Narayandas, N. Long-Term Manufacturer-Supplier Relation-

ships: Do They Pay off for Supplier Firms? The Journal of Marketing 59, 1 (1995),

1–16.

[20] Kamal, F., and Tang, H. Relational Trade: Evidence from the United States. Mimeo,

2015.

[21] Kukharskyy, B. Quality of Institutions and Global Sourcing. Mimeo, 2015.

[22] Kukharskyy, B., and Pfluger, M. Relational Contracts and the Economic Well-

Being of Nations. Discussion Paper, BGPE, 2011.

[23] Macchiavello, R., and Morjaria, A. Competition and Relational Contracts: Evi-

dence from Rwanda’s Coffee Mills. Working Paper, 2014.

21

[24] Manova, K., and Zhang, Z. Export Prices Across Firms and Destinations. The

Quarterly Journal of Economics 127 (2012), 379–436.

[25] Melitz, M. J., and Redding, S. J. Heterogeneous Firms and Trade. Handbook of

International Economics 4 (2014), 1–54.

[26] Monarch, R. It’s Not You, It’s Me: Breakups in U.S.-China Trade Relationships. Work-

ing paper, 2015.

[27] Nunn, N. Relationship-Specificity, Incomplete Contracts, and the Pattern of Trade. The

Quarterly Journal of Economics 122, 2 (2007), 569–600.

[28] Rauch, J. E. Networks versus Markets in International Trade. Journal of International

Economics 48, 1 (1999), 7–35.

[29] Rauch, J. E., and Watson, J. Starting Small in an Unfamiliar Environment. Inter-

national Journal of Industrial Organization 21, 7 (2003), 1021–1042.

22

CENTRE FOR ECONOMIC PERFORMANCE Recent Discussion Papers

1514 Philippe Aghion Antonin Bergeaud Timo Boppart Peter J. Klenow Huiyu Li

Missing Growth from Creative Destruction

1513 Stephen J. Redding David E. Weinstein

Aggregating from Micro to Macro Patterns of Trade

1512 Andrew E. Clark Tom Lee

Early-Life Correlates of Later-Life Well-Being: Evidence from the Wisconsin Longitudinal Study

1511 Clément Bosquet Pierre-Philippe Combes Cecilia García-Peñalosa

Gender and Promotions: Evidence from Academic Economists in France

1510 Andrea Ariu Holger Breinlich Gregory Corcos Giordano Mion

The Interconnections between Services and Goods Trade at the Firm-Level

1509 Johannes Boehm Swati Dhingra John Morrow

Product Diversification in Indian Manufacturing

1508 Hanwei Huang Jiandong Ju Vivian Z. Yue

Structural Adjustments and International Trade: Theory and Evidence from China

1507 Laura Alfaro Pol Antràs Davin Chor Paolo Conconi

Internalizing Global Value Chains: A Firm-Level Analysis

1506 Alessandro Iaria Carlo Schwarz Fabian Waldinger

Frontier Knowledge and Scientific Production: Evidence from the Collapse of International Science

1505 Fabrice Defever Alejandro Riaño

Twin Peaks

1504 Gene M. Grossman Elhanan Helpman Ezra Oberfield Thomas Sampson

The Productivity Slowdown and the Declining Labor Share: A Neoclassical Exploration

1503 Camille Landais Arash Nekoei Peter Nilsson David Seim Johannes Spinnewijn

Risk-Based Selection in Unemployment Insurance: Evidence and Implications

1502 Swati Dhingra John Morrow

Efficiency in Large Markets with Firm Heterogeneity

1501 Richard Murphy Judith Scott-Clayton Gill Wyness

The End of Free College in England: Implications for Quality, Enrolments and Equity

1500 Nicholas Bloom Renata Lemos Raffaella Sadun John Van Reenen

Healthy Business? Managerial Education and Management in Healthcare

1499 Thomas Sampson Brexit: The Economics of International Disintegration

1498 Saul Estrin Daniel Gozman Susanna Khavul

Equity Crowdfunding and Early Stage Entrepreneurial Finance: Damaging or Disruptive?

1497 Francesco Caselli Alan Manning

Robot Arithmetic: Can New Technology Harm All Workers or the Average Worker?

1496 Nicholas Bloom Charles I Jones John Van Reenen Michael Webb

Are Ideas Getting Harder to Find?

The Centre for Economic Performance Publications Unit Tel: +44 (0)20 7955 7673 Email info@cep.lse.ac.uk Website: http://cep.lse.ac.uk Twitter: @CEP_LSE